Real-time inversion of radioactive source distribution using air dose rate measurements via least absolute shrinkage and selection operator method

Shi, W.*; Machida, Masahiko  ; Okamoto, Koji*; Luo, X.*; Feng, W.*; Liu, X.*

The reliability of emergency response in severe nuclear accidents critically depends on robust real-time monitoring of radioactive source distributions. However, this safety function is challenged by physical constraints that create monitoring blind spots and by the inadequacy of static methods in tracking dynamic releases. To enhance the reliability and robustness of source term estimation, this study proposes a dynamic reconstruction framework based on LASSO regression with temporal regularization. A sliding-window time-penalty mechanism is introduced, imposing -norm constraints on inter-step source variations to ensure physical continuity. The contribution matrix and measurement vector are normalized to counteract biases from radiation shielding and time-varying intensities. Validation using a two-room model with internal shielding, with PHITS Monte Carlo simulation, demonstrates accurate reconstruction of dynamic sources from remote measurements. Temporal regularization enhances situational awareness by suppressing spatial aliasing: at sliding-window width (no regularization), hotspot locations fluctuate significantly, with quantitative mean absolute error fluctuations at around , whereas yields improved spatial consistency and the fluctuation quantities decrease to the range. Comparative analysis identifies as optimal in balancing accuracy and computational cost. This work establishes a more reliable pathway for dynamic hazard assessment, enabling accurate localization and intensity tracking under challenging conditions. The proposed framework provides a decision-support tool enhancing the resilience and safety of emergency management in nuclear facilities.